Wolfgang,
> This is what the last paragraph of the documentation you cited wanted to > explain: > ------------------------- > Despite the fact that it would seem to be an obvious win, setting the > preset_nonzero_locations flag to true doesn't seem to accelerate program. > Rather on the contrary, it seems to be able to slow down entire programs > somewhat. This is suprising, since we can use efficient function calls into > PETSc that allow to create multiple entries at once; nevertheless, given the > fact that it is inefficient, the respective flag has a default value equal to > false. > ------------------------ > > This was written when PETSc 2.2 was current, so it may have changed in the > meantime. If you think the default value should be changed, would you mind > removing this hunk of documentation? Now I looked into the function once again and found that we fixed the PETSc matrix structure with MatSetOption (matrix, MAT_NO_NEW_NONZERO_LOCATIONS); in the wrong place. Now the problem is fixed. I then decided to set the default value true because that is the value we also use for the PETSc parallel matrix. However, it would be good if someone tested these two options on the PETSc 3.0 in order to find out whether this is really a good idea. Best, Martin _______________________________________________ dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii
